This study presents a comparative analysis of two deep learning architectures-a Physics-Informed Neural Network (PINN)-Convolutional Neural Network (CNN) hybrid model and a CNN-Only model-for the classification of various cyber intrusions, including Man-in-the-Middle, Quantum Attack, and Side Channel threats. The primary objective was to evaluate whether the integration of a PINN component enhances the detection performance, stability, and generalization capability of the conventional CNN approach. The results demonstrate that the PINN-CNN hybrid model achieved a higher overall test accuracy of 0.9750, marginally outperforming the CNN-Only model's accuracy of 0.9650. Crucially, the hybrid model exhibited superior robustness and generalization, evidenced by the close convergence of its training and validation loss curves, which effectively mitigates the overfitting observed in the CNN-Only model. Furthermore, the PINN-CNN hybrid achieved perfect recall (1.00) for critical attacks such as Man-in-the-Middle and Side Channel, alongside a significantly higher f1-score (0.98 vs. 0.94) for Quantum Attack detection. This work concludes that incorporating physics-informed regularization into the deep learning framework yields a more stable and reliable intrusion detection system, marking a significant advancement in developing robust cyber defenses capable of accurately identifying evolving network threats.
This study presents a comparative analysis of two deep learning architectures-a Physics-Informed Neural Network (PINN)-Convolutional Neural Network (CNN) hybrid model and a CNN-Only model-for the classification of various cyber intrusions, including Man-in-the-Middle, Quantum Attack, and Side Channel threats. The primary objective was to evaluate whether the integration of a PINN component enhances the detection performance, stability, and generalization capability of the conventional CNN approach. The results demonstrate that the PINN-CNN hybrid model achieved a higher overall test accuracy of 0.9750, marginally outperforming the CNN-Only model's accuracy of 0.9650. Crucially, the hybrid model exhibited superior robustness and generalization, evidenced by the close convergence of its training and validation loss curves, which effectively mitigates the overfitting observed in the CNN-Only model. Furthermore, the PINN-CNN hybrid achieved perfect recall (1.00) for critical attacks such as Man-in-the-Middle and Side Channel, alongside a significantly higher f1-score (0.98 vs. 0.94) for Quantum Attack detection. This work concludes that incorporating physics-informed regularization into the deep learning framework yields a more stable and reliable intrusion detection system, marking a significant advancement in developing robust cyber defenses capable of accurately identifying evolving network threats.
The rapid expansion of the Internet of Things in critical industrial environments has significantly increased the attack surface, exposing systems to sophisticated cyber threats. Traditional pattern-based intrusion detection systems struggle to detect such advanced attacks, while deep learning approaches, despite achieving high detection accuracy, often suffer from high computational cost and latency, limiting their deployment in resource-constrained edge gateways. Quantum machine learning offers a promising alternative by enabling high-dimensional feature representation; however, current implementations are constrained by hardware noise, limited qubit availability and backend-dependent execution characteristics in the Noisy Intermediate-Scale Quantum era. To address these challenges, this paper proposes a Residual Hybrid Quantum-Classical Neural Network (RHQ-CNN) for efficient intrusion detection in Edge-Industrial Internet of Things (Edge-IIoT) environments under idealized quantum simulation conditions. The proposed framework employs a classical neural network encoder to compress high-dimensional network traffic into a compact latent representation suitable for quantum processing. A variational quantum circuit with serial data re-uploading is then utilised to model complex non-linear decision boundaries without increasing qubit requirements. In addition, a residual connection fuses classical and quantum representations to improve training stability and preserve latent feature information. The model is evaluated on the Edge-IIoTset dataset and achieves a test accuracy of 99.94%, with high weighted performance across the 15-class detection task, although the extremely low-sample Fingerprinting class remains comparatively more challenging. Additional controlled ablation, deployment-cost, and bootstrap confidence interval analyses demonstrate the test-set metric stability and computational trade-offs of the proposed architecture. These findings highlight the potential of hybrid quantum-classical models for next-generation cybersecurity in industrial IoT systems.
In this work, we investigate the performance of hybrid quantum-classical convolutional neural networks (QCNNs) in medical image classification under the presence of adversarial attacks. Adversarial attacks pose a significant threat to image classification models, since they introduce subtle perturbations to input images, leading to incorrect predictions. Specifically, in this work, 10 QCNNs with their corresponding classic Convolutional Neural Networks (CNNs) counterparts are comparatively examined for their robustness, on two different medical image datasets using transfer learning, under two different adversarial attacks. In a subsequent step, the implementation of training models from scratch is also evaluated. With transfer learning on the X-ray dataset, CNN and QCNN models achieve high and comparable performance across most backbones. On the histopathological dataset, CNNs and QCNNs often generalize worse compared to the X-ray dataset, indicating head overfitting under domain shift. The adversarial evaluation results highlight that most models' robustness comes from the backbone rather than the head. QCNN models perform comparably to CNNs under attack and show improvements when trained from scratch. Overall, the results indicate that integrating quantum circuit mechanisms into classical CNN architectures can enhance the robustness of models in medical imaging tasks, representing a promising domain that warrants further investigation.
The objective of this research is to develop a Hybrid Quantum-Classical Convolutional Neural Network (QC-CNN) framework that integrates parameterized quantum circuits (PQCs) within classical CNN architectures to enhance image classification performance.The proposed model leverages quantum principles such as superposition and entanglement for high-dimensional feature representation, achieving superior accuracy and reduced computational complexity.Experiments on MNIST and CIFAR-10 datasets demonstrated improved classification accuracy of 98.7% and 82.5%, respectively, surpassing traditional CNNs while reducing training time by 33% and parameters by 45%.Statistical analysis confirmed the significance of these improvements.Visualizations using t-SNE revealed enhanced class separability, and noise perturbation tests validated the model's robustness.The results highlight the hybrid QC-CNN's potential for efficient and scalable quantum-enhanced deep learning applications.Extended evaluations, including quantum-layer-depth analysis, robustness testing, and t-SNE visualization, empirically support the hybrid QC-CNN's effectiveness across varied scenarios.
With an increasing sophistication and prevalence of cyber attacks, basic machine learning techniques have been shown to lack basic capabilities for network attack detection in complex, high-dimensional attacks. In this paper, the authors present and thoroughly test a hybrid Quantum Machine Learning (QML) approach to facilitate the detection and classification of advanced cybersecurity threats. The QETC framework uses quantum feature mapping and data encoding with entanglement to increase the representational power of high dimensional data over classical methods. Systematic evaluation was performed on the NSL-KDD and CICIDS-2017 benchmark intrusion detection datasets with a simulated quantum backend with 8 qubits. Experimental results show that the QETC model obtains classification accuracy of 96.3% and 97.1% on the NSL-KDD and CICIDS-2017 datasets, respectively, which is statistically significant, with a margin of 2.8 and 3.4 percentage points over the best-performing classical baseline models. Interestingly, the quantum model outperformed by a large margin the detection capabilities for both rare and zeroday attack classes with an F1-score improvement up to 8.2 percentage points in the minority attack classes. This result in this paper demonstrates significant empirical evidence that quantum-classical hybrid architectures are a promising and viable path forward for next-generation cybersecurity applications.
Weaknesses arising from heterogeneous device structure, high-dimensional traffic patterns, and more complex nonlinear attack behaviors. The traditional intrusion detection systems (IDS), which rely on the classical machine learning models, often face the challenges of scalability, feature explosion, and low representational ability when handling large-scale IoT traffic data. To meet these challenges, we propose a Hybrid Quantum-Classical Variational Neural Network (HQCVNN) for effective intrusion detection in IoT systems. The model utilizes a Variational Quantum Circuit (VQC) as a nonlinear quantum feature map followed by dense neural layers for overall classification. Angle embedding with entanglement layers enables the encoding of quantum data and leads to a better separability of features in high-dimensional space, whilst the classical optimization guarantees efficient training on NISQ time scales. The model is implemented using Qiskit and evaluated on classical benchmark intrusion detection datasets. Experimental results show the proposed method has better classification accuracy, F1-score, and complexity reduction than conventional ANNs and SVMs. The results demonstrate the promise of hybrid quantum machine learning systems with end-to-end general-purpose capabilities in future IoT security and provide pragmatic impressions on how to deploy them on near-term quantum processors.
This study presents a systematic comparison between hybrid quantum-classical neural networks and purely classical models across three benchmark datasets (MNIST, CIFAR100, and STL10) to evaluate their performance, efficiency, and robustness. The hybrid models integrate parameterized quantum circuits with classical deep learning architectures, while the classical counterparts use conventional convolutional neural networks (CNNs). Experiments were conducted over 50 training epochs for each dataset, with evaluations on validation accuracy, test accuracy, training time, computational resource usage, and adversarial robustness (tested with ε=0.1 perturbations).Key findings demonstrate that hybrid models consistently outperform classical models in final accuracy, achieving {99.38\% (MNIST), 41.69\% (CIFAR100), and 74.05\% (STL10) validation accuracy, compared to classical benchmarks of 98.21\%, 32.25\%, and 63.76\%, respectively. Notably, the hybrid advantage scales with dataset complexity, showing the most significant gains on CIFAR100 (+9.44\%) and STL10 (+10.29\%). Hybrid models also train 5--12× faster (e.g., 21.23s vs. 108.44s per epoch on MNIST) and use 6--32\% fewer parameters} while maintaining superior generalization to unseen test data.Adversarial robustness tests reveal that hybrid models are significantly more resilient on simpler datasets (e.g., 45.27\% robust accuracy on MNIST vs. 10.80\% for classical) but show comparable fragility on complex datasets like CIFAR100 ($\sim$1\% robustness for both). Resource efficiency analyses indicate that hybrid models consume less memory (4--5GB vs. 5--6GB for classical) and lower CPU utilization (9.5\% vs. 23.2\% on average).These results suggest that hybrid quantum-classical architectures offer compelling advantages in accuracy, training efficiency, and parameter scalability, particularly for complex vision tasks.